sqlx-sqlite 0.7.2 didn't get picked up

[abonander]

The docs for sqlx-sqlite 0.7.2 didn't get built, but it doesn't appear at all in the build queue or in the recent failures list: https://docs.rs/sqlx-sqlite/0.7.2

The strange part is, it appears that the rest of the crates got picked up even though I published them all at basically the same time: https://docs.rs/sqlx/0.7.2/sqlx/index.html

[Nemo157]

Another instance of #797

In this case, I assume the issue will be the circular exact version dev-depencency on sqlx, the queue is too fast now and attempted to build sqlx-sqlite before sqlx was published. A rebuild should succeed (I don't have access right now to queue it up, hopefully @syphar can tomorrow).

We should maybe see if there's some way we could avoid this, there's the brute-force "delete [dev-dependencies] from Cargo.toml" but maybe there's something cleaner we could do (and I'm not sure how that interacts with --scrape-examples).

[syphar]

I queued the rebuild.

This is definitely an unexpected side-effect of parallel builds, perhaps we should extend our build-attemts system to help here. I remember that we don't always use it & retry? @Nemo157

[abonander]

Is there anything we can do on our end to mitigate this in the future? The circular dependency is kinda necessary in order to have compile-tested examples using the facade in these lower APIs, though I suppose we could make a fake facade just for the examples.

[syphar]

for now the parallel builds are stopped until we solved another issue.

After that I believe that you won't be the only case, so docs.rs should have a way to handle these cases.

Otherwise, the solution would probably be to only release sqlx-sqlite when the needed version of sqlx is already on crates.io. ( assuming the error is what was decribed above)

[abonander]

The problem is that sqlx of course depends on sqlx-sqlite since the latter is included in the facade of the former.

[syphar]

in any case, the (simplest) solution on docs.rs side stays the same: retry the build :)

[syphar]

( apart from fixing #797)

[syphar]

After having talked to @Nemo157 I'll close this issue, while creating #2262 for the proposed solution.

While the publisher of the crate could prevent this problem by first releasing the base-crate, and then the dependants, we believe this should be handled by docs.rs.
To be exact: we already have a thing that handles this: we are retrying failed builds.

But our current retry implementation has the edge-case that it retries too quickly if the queue is empty.

So we will add a delay to these: #2262

Also, #797 is still unsolved.